JPH04212600A - Voice input device - Google Patents

Abstract

PURPOSE:To present the voice input device to select only an audio signal from a sound source at a desired position by using plural directional microphones. CONSTITUTION:The power of audio signals sound-collected at microphones A1-An is measured by power measuring instruments B1-Bn, next, an SNR is measured by SNR measuring instruments C1-Cn and at a comparator circuit 11, it is judged by using the conditional expression of Smax (maximum value)/ Smin (minimum value)<1+Th (threshold value) or the like whether the audio signal comes from the sound source at the desired position from power S1-Sn or not. When it is judged that the audio signal comes from the sound source at the desired position, a selection enable signal is supplied to a selecting circuit 12. When the selection enable signal is inputted, the selecting circuit 12 detects the maximum value of SNR information or power information, selects the audio signal of the microphone corresponding to the maximum value and outputs the signal to a voice recognizing device 2.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice input device that uses a plurality of directional microphones, selects an optimal voice signal from among the microphones, and supplies the voice signal to a voice recognition device or the like.

0002

2. Description of the Related Art Conventionally, in order to increase the speech recognition rate of a speech recognition device, it has been required to reduce as much as possible the noise contained in the speech signal supplied from a microphone or the like to the speech input section of the speech recognition device. There is. Therefore, various methods have been considered to remove noise when performing speech recognition. (1) One method is to collect sound waves in a space without noise when converting them into electrical signals using a microphone, etc. (2) The other method is to convert sound waves into electrical signals using a microphone, etc. A possible method is to perform signal processing later and extract only the desired audio signal.

[0003] In the method (1), if a microphone with narrow directivity angle is used, for example, it is easy to obtain only an electric signal that collects only sound waves from a sound source within a specific narrow range.

[0004] In the method (2) above, background noise included around an electrical signal (for example, an audio signal) from a desired sound source is removed by filtering only the desired audio signal using a filter. It's a method.

[0005]

[Problems to be Solved by the Invention] However, the voice input method described above has the following problems. In other words, even when a microphone with a narrow directivity angle is used, noise coming from the direction of the microphone cannot be removed by the method (1) above. Further, in the method (2), it is possible to deal with stationary noise, but it is difficult to deal with the sudden occurrence of noise (random noise).

The present invention has been made in view of the above problems, and its purpose is to provide an audio input device that selects an optimal audio signal and supplies it to the outside.

[0007]

[Means for Solving the Problems] In order to achieve the above object, the present invention aims to improve the power and SNR (signal ratio) of audio signals collected by multiple microphones with narrow directivity angles.
The audio input device was improved so that the microphone that provides the optimal audio signal was selected by measuring the ratio (l to NoiseRatio).

That is, in an audio input device that collects audio signals using a plurality of directional microphones and obtains a desired audio signal from the collected audio signals, the audio signals collected by the plurality of microphones are a determining means for determining whether or not the voice signal is generated from a sound source at a desired position; It is characterized by having a selection means for selecting a desired audio signal.

[0009]

[Operation] According to the present invention, the determining means determines whether or not the sound signals collected by the plurality of directional microphones are produced from a sound source at a desired position, so that the sound signals are produced by a desired speaker. It is determined whether the person has
If it is determined that the audio signal is from a sound source at a desired position, the selection means selects the audio signal having the maximum value of SNR information or power information of the plurality of audio signals, thereby obtaining the optimal audio signal. Noise signals from sources other than those located at are not selected.

0010

DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, a preferred embodiment of the voice input device according to the present invention will be described with reference to the drawings. Identical members in the figures are given the same reference numerals.

FIG. 1 shows a functional block diagram of the voice input device 1. As shown in FIG. In the figure, A1 to An are microphones, B1 to Bn are power measuring devices, C1 to Cn are SNR measuring devices, 11 is a comparison circuit, and 12 is a selection circuit. 2 is a voice recognition device.

In this audio input device 1, microphones A1 to An with narrow directivity are appropriately arranged in space, and each microphone receives an audio signal Si1 as an electric signal.
~Sin and then supplied to the corresponding power measuring devices B1 to Bn, respectively. The power measuring devices B1 to Bn measure the power of the supplied audio signal and the power of noise. If the audio signal is expressed as x(t), the power P is x
2(t). The powers of the audio signals obtained here are respectively S1 to Sn, and the signal when no audio signal is input is noise, and the power of this noise is N1 to Nn. The power of this audio signal S1~
Sn and noise powers N1 to Nn are supplied from the respective power measuring devices B1 to Bn to the SNR measuring devices C1 to Cn and the selection circuit 12.

The SNR measuring devices C1 to Cn calculate the SNR from the input audio signal powers S1 to Sn and the noise powers N1 to Nn. Each SNR measuring device C1-C
SNR1 to SNRn obtained at step n are supplied to the selection circuit 12. The comparison circuit 11 checks whether the power signals S1 to Sn of the input audio signal are emitted from a sound source within a specified region. If it is determined that the sound was emitted in a specified area, a selectable signal is supplied to the selection circuit 12.

Here, the method for determining whether or not the sound is emitted from a sound source within the specified area is performed as follows. That is, among the power signals S1 to Sn of the audio signals supplied from the power measuring devices A1 to An, the maximum power S
max and the minimum power is Smin, then Sm
It is determined whether condition (1) of ax/Smin<1+Th is satisfied. Here, Th represents a threshold value, and is a constant determined by adjusting the spatial arrangement of each microphone, the gain of the microphone, and the like. Note that the above conditions are just one example.

If this condition is not met, it is determined that the sound is not emitted within the directional region of the microphone, and the selectable signal is not supplied to the selection circuit 12. However, if it is determined that the conditions are satisfied, a selectable signal is supplied to the selection circuit 12. The processing in the comparison circuit 11 serves as a preprocessing function for the selection circuit 12, which will be described later, to finally select one of the most suitable audio signals from the microphone.

An example of a circuit configuration for realizing the comparator circuit 11 will be described below with reference to FIG. In the figure, the comparison circuit 11 includes a maximum value detector 111, a minimum value detector 112, a divider 113, and a comparator 114. Power signals S1~ of audio signals supplied from SNR measuring devices C1~Cn
Sn is the maximum value detector 111 and the minimum value detector 11, respectively.
2, and the maximum values Smax and Smin are stored. Next, the divider 113 divides Smax and Smin, and the comparator 114 compares whether the division result is smaller than the value (1+Th) input from the outside,
If it is small, a selectable signal is output, and if it is not small, a selectable signal is not output.

The above condition (1) supplied to the selection circuit 12
The power signal Si and its SN of multiple audio signals that satisfy
One piece of information about R is selected here. The selection method is as follows. That is, the power signal Si of the microphone audio signal exhibiting the highest SNR value among the input SNRs is selected.

An example of a circuit configuration for realizing the selection circuit 12 will now be described with reference to FIG. In the figure, selection circuit 1
2 is maximum value detector 121, 122 and selection switch 12
It consists of 3. The SNR1 to SNRn information supplied from the SNR measuring devices C1 to Cn is supplied to the maximum value detector 122, and the maximum SNR is selected. Further, the power signals P1 to Pn supplied from the power measuring devices B1 to Bn are also supplied to the maximum value detector 121, and the maximum power signal is selected. Next, if the selectable signal is supplied from the comparison circuit 11 to the selection switch 123, the maximum SN
The audio signal for which R can be obtained is selected from among Si1 to Sin and output.

The audio signal selected here is finally output as the output signal of this audio input device 1, and this output signal is supplied to, for example, the audio input section of the audio recognition device 2, and when the audio is recognized, it is faster than conventional audio signals. The speech recognition rate can be increased by suppressing the influence of noise. In this way, an audio signal with a good SNR can be automatically selected from audio signals collected by a plurality of microphones.

Furthermore, in the above embodiment, the value of SNR was used as an index for selecting the optimum audio signal, but more simply, the value of the power P of the input audio signal is compared instead of the SNR, and the value of the power P of the input audio signal is compared. In some cases, it is also possible to select the optimal audio signal by selecting the highest power. By doing this, the SNR measuring devices C1 to C
Since the provision of n can be omitted, the scale of the hardware can be reduced.

Next, FIG. 4 shows how to actually perform voice recognition using the voice input device 1 using two microphones. In the figure, user C is sitting in front of a personal computer 3, and two microphones A1 and A2 are installed in front of user C on the left and right to collect the words spoken by user C, and the collected audio signal is is processed by the personal computer 3 via the voice input device 1 and the voice recognition device 2, and displayed on the display of the personal computer 3 by the user C.
This is an operational diagram in which a sentence is created by displaying the words spoken by a person as a string of characters.

The directivity angle of microphone A1 is θ1, the directivity angle of microphone A2 is θ2, and the range where the directivity range of microphone A1 and the directivity range of microphone A2 intersect is defined as intersection range D (shaded range), a noise source A exists within the directional range of microphone A1, and this noise source A is located behind user C, and a noise source B exists within the directional range of microphone A2. It is assumed that the noise source B is located behind the user C.

The operation of the voice input device 1 will be explained based on the device arrangement as described above. The case where only noise source A exists and generates noise in FIG. 4 will be described. Noise source A is located within the directivity range of microphone A1, but not within the directivity range of microphone A2. Microphone A
Let the power of the noise collected by microphone A2 and measured by power measuring device B1 be N1, and the power of the voice emitted by user C be S1, and the power of the noise collected by microphone A2 and measured by power measuring device B2. If the power is N2 and the power of the voice emitted by the user C is S2, the noise source A is naturally located within the directivity range of the microphone A1, so the noise power N1>N2.

Therefore, the relationship between SNR1 of the audio signal collected by microphone A1 and SNR2 of the audio signal collected by microphone A2 is SNR1 (=S1/N1)
<SNR2 (=S2/N2). In this way, in the selection circuit 12, the voice signal collected by the microphone A2 is supplied to the voice recognition device 2, and the noise signal from the noise source A is not mixed into the voice recognition device, and the voice signal is recognized by the noise. No reduction in rate.

Next, the operation of the voice input device 1 when the noise source B suddenly emits noise will be explained. Noise source B is located within the directivity range of microphone A2, and
It is not located within the directivity range of 1. In this case, the power of the noise becomes N1<N2, and the power of the noise collected by the microphone A2 becomes larger.

Therefore, SNR1>SNR2, and the selection circuit 12 supplies the voice signal collected by the microphone A1 to the voice recognition device 2.

Next, this embodiment describes a method for preventing malfunctions in the case where the noise source B is a human being and the voice recognition device malfunctions due to the voice spoken by the noise source B instead of the user C speaking. The voice input device 1 will be explained.

In order to prevent the above-mentioned malfunction, it is determined whether or not the noise source B exists in a specified area (in FIG. 4, the intersecting range D between the directivity of the microphone A1 and the directivity of the microphone A2). If it is determined that the audio signal exists within the specified area, the audio signal is selected.However, as shown in FIG. The emitted voice signal is not supplied to the voice recognition device 2.

Specifically, the determination is made as follows. In other words, the microphones A1 and A2 collect sound, and the collected audio signals are measured by the power measuring devices B1 and B2, respectively.The SNR measuring devices C1 and C2 then calculate the SNR, and in the SNR measuring device C1, the SNR1= S1/N
Get 1. In addition, in the SNR measuring device C2, SNR2=
Obtain S2/N2. These SNR information are supplied to the selection circuit 12. In the comparison circuit 11, the following conditions are met.
By modifying the condition 1, for example, 1-Th < S1/
S2<1+Th, but 0<Th<<1, e.g. Th
=0.1 (this is set as condition (2)). Check whether this condition is satisfied.

That is, in this example, microphone A
Since microphones A1 and A2 are located at the same distance from user C, if there is no noise source B, the power of user C's audio signal at microphones A1 and A2 is considered to be approximately equal, so S1/S2 is approximately 1. considered to be equal. Therefore, a selectable signal that satisfies the condition (2) is supplied to the selection circuit 12, and the selection circuit 12 detects the maximum value of the SNR information supplied from the SNR measuring devices C1 and C2, and selects the corresponding microphone. The input voice of A1 or the input voice of the microphone A2 is selected.

However, in FIG. 4, the noise source B is located outside the intersection range D, so the power of the audio signal at the microphone A2 due to the audio emitted from the noise source B is set as S2, and the power of the audio signal from the noise source B at the microphone A1 is set as S2. When the power of the audio signal is S1, the relationship between S1 and S2 is S1<<S2. This is because the noise source B is located outside the directivity range of the microphone A1, so the audio signal emitted from the noise source B is not collected by the microphone A1 and is therefore close to zero. Therefore, the above relationship S1<<S
2, and this relationship does not satisfy the condition (2). Therefore, since the comparator circuit 11 does not supply a selectable signal to the selection circuit 12 depending on the audio signal from the noise source B, no audio signal is output from the noise source B to the speech recognition device 2. If the noise source does not exist within the directivity range of either microphone A1 or A2, there is a possibility that the above condition (2) is satisfied, but even if it is satisfied, it is outside the directivity range, so Microphone A
Since the power of the audio signals collected by A1 and A2 is considered to be close to 0, there is no risk of malfunctioning of the audio recognition. In this way, the user can be placed within the intersecting range of the directivity of the two directional microphones, and can be prevented from selecting audio signals from noise sources outside the intersecting range of the directivity. Therefore, such voice input devices can be used as voice input units for voice recognition devices and acoustic devices installed in plants that manufacture various products, aircraft, vehicles, etc., and can contribute to improving voice recognition rates. I can do it.

[0032] In the above embodiments, the voice input device for supplying voice signals to a voice recognition device was explained as an example, but the present invention is not limited to the voice recognition device, and may be applied to other audio devices as well. This voice input device can be applied.

[0033]

[Effects of the Invention] As described above, the present invention provides the following effects.

In other words, by having the judgment means and the selection means, only the audio signal from the sound source at the desired position is selected with the maximum value of the power information or SNR information of the audio signal, and the noise from the sound source other than the desired position is selected. Since it is not selected, an optimal audio signal can be obtained.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of a voice input device according to this embodiment.

FIG. 2 is a circuit diagram of a comparison circuit in FIG. 1;

FIG. 3 is a circuit diagram of a selection circuit in FIG. 1;

FIG. 4 is a conceptual diagram of an example of operation of the voice input device.

[Explanation of symbols]

1 Voice input device 11 Comparison circuit (judgment means) 12 Selection circuit (selection means) A1~An Microphone B1~Bn power measuring device C1~Cn SNR measuring device

Claims (1)

[Claims] 1. An audio input device that collects audio signals using a plurality of directional microphones and obtains a desired audio signal from the collected audio signals, the audio signal collected by the plurality of microphones. determining means for determining whether or not the sound signal is generated from a sound source at a desired position; An audio input device comprising a selection means for selecting a desired audio signal based on a value.